JPH04127721A - Acoustic echo canceler - Google Patents

Abstract

PURPOSE:To improve the performance and reliability of the echo canceler by judging whether in a double talk state or an echo bus change stage and executing optimum control corresponding to the respective states. CONSTITUTION:An adaptive digital filter (ADF) control circuit 21 is composed of a prestage tap coefficient square sum circuit part 22 as a prestage power detecting means, poststage tap coefficient square sum circuit part 23 as a poststage power detecting means and decision circuit 24 as a comparing and deciding means to control an ADF 11 by comparing signals outputted from these circuit parts 22 and 23. Since it is decided by comparing the sizes of residual power in pre-and post-stages whether the state is the double talk state or not, either the double talk state or the echo bus change state can be accurately judged, and exact adaptive control can be executed by the ADF 11. Thus, the performance and reliability of the acoustic echo canceler can be improved.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] This invention relates to an acoustic echo canceller that cancels acoustic echo by determining whether it is in a double talk state or in an echo path fluctuation state.

[Prior Art] Generally, acoustic echo cancellers are used to cancel acoustic echoes caused by acoustic coupling between speakers and microphones in hands-free calls such as voice communication conference systems. As this acoustic echo canceller, for example, the one described in ``A Method for Detecting Superimposed Speech in an Adaptive Echo Canceller'' is known. This acoustic echo canceller will be explained based on FIG. 2.

The audio signal from the other party passes through the reception terminal 1 and is converted into a digital signal X(k) (hereinafter, k represents the sampling time) by the A/D converter 2. The signal is converted into an analog signal by the /A converter 3, and is output from the speaker 4 to the near-end speaker A as the voice from the other party.

Reference numeral 5 denotes a microphone, and the voice emitted from the near-end speaker A is captured by this microphone 5, and converted into a digital signal y (k) by an A/D converter 6 as an audio signal s (k), and processed by each part. The signal is then converted into an analog signal by a D/A converter 7, and transmitted to the other party (far-end party) via a transmission terminal 8.

11 is an adaptive digital filter (hereinafter referred to as [ADF]); 12 is an adder provided on the transmission signal line 13;
The FII forms a pseudo echo signal V (k) based on the digital signal x (k) from the receiving side, and outputs this pseudo echo signal y (k) to the adder 12 . Adder 12
, a pseudo echo signal y is obtained from a digital signal y(k) which is an audio echo signal input from the microphone 5 side.
(k) is subtracted.

ADFII specifically has the configuration shown in FIG.

In the figure, T, ~TN are unit delay elements, which are provided for the set number of taps, and are connected to the digital signal x(k
) is done manually. B o-B N is a tap coefficient generator;
Each tap coefficient generator Bo-BN generates a residual signal e (
k) is set by feeding back the filter tap coefficient. , hl...hL...hM...hN. co-CN is a multiplier, and in this multiplier C8 to CN, the digital signal x from each unit delay element T1 to TN is
(ki) and each tap coefficient generator B. ~Filter tap coefficients from BN. ,h,...hL...h□・-・
hN and added by an adder to obtain a pseudo echo signal y(k)=Σhi(k)・x(ki)−(1
) is formed as.

Further, the tap coefficient h1(k) is a transfer characteristic between the speaker 4 and the microphone 5, that is, an impulse response,
In order to follow changes in the acoustic coupling state (hereinafter referred to as "echo bass"), that is, to minimize the residual signal e (k), adaptive control is applied to boat fishing using the learning identification method shown in the following equation. It will be done.

e(k)・X(ki-i) hH(k+1) = hl(k)+α・
...(2) Σ x2(k-i) α: Loop gain (correction coefficient) In addition, in FIG. update)
This is provided to prevent the person from pointing in the wrong direction. In other words, when near-end speaker A makes a sound and enters a double talk state, the residual signal e(k) appears to increase, and it is not true that the coefficients in equation (2) are updated as is. It is no longer possible to estimate the echo path of the acoustic echo, and as a result, the original acoustic echo cannot be canceled. Therefore, the double talk detector 15 detects the signal from the near-end speaker A as soon as possible and stops updating equation (2).

Specifically, this double talk detector 15 starts from the level Lx of the digital signal x (k) at one example of the receiving terminal A, which will be described later. Estimate or, l, and A from Lx, which will be described later. . □
By subtracting the estimated residual level L, output A().
The estimated residual level L of the ACOM estimation circuit 16 is based on the residual level L RES of the residual signal e (k) and the level Lx of the digital signal x (k).
A correction circuit 17 for correcting RES and a residual level LRE
The comparison circuit 18 compares S and the estimated residual level L RES and controls whether to stop updating the equation (2) or continue depending on the difference in magnitude.

Here, the residual level L RES is the level of the residual signal e(k) when there is no voice signal s (k) from the near-end speaker A. Furthermore, as described above, the estimated residual level L RES is obtained by subtracting A.OM from the level Lx of the digital signal x (k) in the ACOl''l estimation circuit 16. That is, L RES = L A C (obtained by IH. It is a value obtained by Aco-1 (Aecho + Acan), where A echo is the amount of acoustic coupling between the speaker 4 and the microphone 5, and Acan is the amount of echo cancellation canceled by the converter 12. be.

Then, the double talk detector 15 detects the residual level L R
The comparison circuit 18 compares ES and the estimated residual level LRE'3, and when LRES>LRβ, it is determined that there is a double talk state, and the update of the coefficient of ADFII, that is, the update of equation (2) is stopped. is set to .

In the acoustic echo canceller configured as described above, when the voice from the other party is emitted from the speaker 4, a part of the voice is inputted from the microphone 5 as an acoustic echo. This signal is converted into a digital signal y (k) by the A/D converter 6 and input to the adder 12 . on the other hand,
In ADFII, a pseudo echo signal y (k) is generated using the digital signal x (k) on the reception terminal 1 side as a reference signal.
is formed, the acoustic echo is canceled by subtracting this pseudo echo signal y (k) from the echo signal y (k), and the resulting residual signal e (k) is converted into an analog signal by the D/A converter 7. The signal is converted and sent from the transmission terminal 8 to the far-end speaker.

Furthermore, in the call state, the double talk detector 15 constantly detects whether or not there is a double talk state.
When a double talk state is determined under the above-mentioned conditions, coefficient updating is stopped, and when a single talk state is determined, coefficient update control is performed.

[Problems to be Solved by the Invention] However, in the conventional acoustic echo canceller described above, even when the near-end speaker A does not make a sound, when the volume of the speaker 4 is increased or when an echo bus fluctuation occurs, the L RES is > L RES conditions may be met and a double talk state may be erroneously determined. In this case, there is a problem in that the coefficients are not updated, Acan decreases, and the acoustic echo returns to the far-end speaker without being reduced, resulting in poor communication performance.

Furthermore, when the acoustic echo becomes extremely large, the loop gain increases and so-called howling occurs, making it impossible to make a call.

This invention was made in view of the above problems, and
It is an object of the present invention to provide an acoustic echo canceller that can accurately judge whether it is a double talk state or an echo path fluctuation state and perform optimal control according to each state.

[Means for Solving the Problems] The present invention has been made in view of the above-mentioned problems, and it combines a pseudo echo signal and an acoustic echo signal to remove reverberant echoes, and also removes reverberant echoes by using an adaptive filter when in a double talk state. Applied to acoustic echo cancellers that stop adaptive control.

In such an acoustic echo canceller, a front-stage power detection means detects the magnitude of the previous stage out of the residual power after the acoustic echo is removed, a rear-stage power detection means detects the magnitude of the latter stage, and each power detection means The present invention is characterized by comprising a comparison and determination means that compares the output values from the means and determines that a double talk state exists when the increase rate of the residual power in the latter stage is large.

[Function] With the above configuration, the front-stage power detection means always detects the magnitude of the residual power in the former stage, and the rear-stage power detection means always detects the magnitude of the residual power in the latter stage.

The respective values detected by each power detection means are compared by a comparison and determination means, and it is determined which one is larger. If the rate of increase in the latter stage of the residual power is large, it is determined that a double talk state exists.

[Example 1] An example of the present invention will be described below with reference to FIGS. 1 and 4(a),
The explanation will be based on (b). The overall configuration of the acoustic echo canceler of this embodiment is almost the same as that of the conventional acoustic echo canceller described above, so the same members are given the same reference numerals and the explanation thereof will be omitted here.

FIG. 1 is a block diagram showing the configuration of the acoustic echo canceller of this embodiment, FIG. FIG. 4(b) is a characteristic diagram showing a residual power curve corresponding to the impulse response of FIG. 4(a).

First, the characteristics of the impulse response and the corresponding residual power curve will be explained based on FIGS. 4(a) and 4(b).

As shown in FIG. 4(a), the impulse response characteristic line has a large initial amplitude after the flat delay a, which corresponds to the propagation delay between the speaker 4 and the microphone 5.
It generally has the property of gradually attenuating. This property does not change much even with changes in the echo path, and shows a similar tendency. Since the residual power curve corresponds to the impulse response, that is, the filter tap coefficient hi (k) of ADFII has a one-to-one correspondence with the sample sequence of this impulse response, so the result shown in Fig. 4(b) is as follows. As shown, it has almost the same characteristics as the characteristic line in FIG. 4(a). In other words, it has a characteristic of having a large initial value and gradually attenuating.

On the other hand, the tap coefficient h1(k) is updated by adaptive control in response to variations in the echo path. Furthermore, it is also updated when the near-end speaker A's voice signal s (k) is input. That is, by inputting the audio signal s (k), the residual signal e (k
) increases, and the tap coefficient h
H(k) is updated and the residual power increases.

In other words, the residual power does not change much with respect to changes in the echo path, but changes sharply when the near-end speaker A's voice signal s (k) is input (double talk state occurs). Specifically, the tap coefficient hH(k) becomes a value far from the appropriate value, and the value of the residual power becomes disordered. In particular, Figure 4 (
It has a characteristic that the first stage C of the tap number region in a) does not change much, but it increases in the latter stage.

Therefore, by monitoring the fluctuations in the residual power at the front stage C and the rear stage C, it is possible to accurately determine whether it is an echo path fluctuation state or a double talk state.

Next, the configuration of the acoustic echo canceller of this embodiment is based on FIG. 1.

The overall configuration of the acoustic echo canceller of this embodiment is almost the same as that of the conventional acoustic echo canceller described above, but in this embodiment, the conventional double talk detector 15 is replaced by A
A DF control circuit 21 is provided.

This ADF control circuit 21 includes a front-stage tap coefficient 2-TOW circuit section 22 as a front-stage power detection means, a rear-stage tap coefficient 2-TOW circuit section 23 as a rear-stage power detection means, and outputs from these circuit sections 22 and 23. Compare the signals and A
Judgment circuit 24 as comparison judgment means for controlling DFII
It is composed of.

The pre-stage tap coefficient 2 Towa circuit section 22 squares and adds up the tap coefficient hi (k) of the pre-stage C in FIG. 4(a), and is based on the following calculation formula. Note that the sum of the squares of the tap coefficients hi (k) is calculated using this tap coefficient hi
This is because (k) has a one-to-one correspondence with the impulse response sample sequence, so the sum of squares of the tap coefficient h1(k) is equivalent to the residual power.

ΣhH(k)=H1...(3) 1 厘0 Post-stage tap coefficient 2 The Towa circuit section 23 squares and adds the post-stage tap coefficient hH(k) in FIG. 4(a). According to the following calculation formula.

Xhick)=H2=(4) Note that H1 is referred to as the front stage power level, and H2 is referred to as the rear stage power level.

The determination circuit 24 determines the front stage power level H1 from the front stage tap coefficient 2 to sum circuit unit 22 and the rear stage power level H2 from the rear stage tap coefficient sum of squares circuit unit 23, compares them with each other, and determines the ADFII as the next stage. Control with conditions.

In a state where both the first stage power level H1 and the second stage power level H2 have small increases and decreases, if Hl>a2H2 a2: dark value, it is determined that the echo path fluctuation state (single talk state) is present.

Further, in a state where the amount of increase in the subsequent power level H2 is large and H1≦a2H2, a double talk state is determined. Note that the increase/decrease in the pre-stage power level H1 may be large or small.

If it is determined that the echo path is in a fluctuating state, the ADFll is controlled to perform adaptive control, and if it is determined to be a double talk state, the adaptive control is stopped, and the tap coefficient immediately before this stop is held and the ADFll is controlled to perform adaptive control. Equation (1) is calculated using the tap coefficients to form a pseudo echo signal y(k).

Furthermore, the determination circuit 24 also has a determination function for leaving the double talk state when it is determined that the state is a double talk state. That is, the set value RES(th) and the residual level L
Compare with RES, L RES(th) > l,,
When RE3 is reached, the determination circuit 24 controls ADFII to exit the double talk state.

Note that the number N of taps of ADFII is determined in consideration of the impulse response length. That is, from FIG. 4(a), the number N of filter taps required to obtain an echo cancellation amount of 30 dB corresponding to a residual power of -30 dB is determined. Here, the numerical value of the amount of echo cancellation (30 dB) is an example and varies depending on each acoustic echo canceller. Furthermore, Ho and H2 for determining whether or not there is a double talk state
threshold value a2 (usually 50 or more), the number of taps in the front stage C, the number N-M of taps in the rear stage, and the set value RES
Since (th) differs depending on the type of acoustic echo canceller, it is set in accordance with the acoustic coupling condition serving as the design model.

Next, the operation of the acoustic echo canceller configured as above will be explained. Note that the overall operation of this acoustic echo canceller is almost the same as that of the conventional acoustic echo canceller described above, so here, the ADF control circuit 21 will be explained.
We will only explain the effect of .

The ADF control circuit 21 is always functioning, and its pre-stage tap coefficient 2 Towa circuit section 22 is the pre-stage C in FIG. 4(a).
The value equivalent to the residual power in the subsequent stage is calculated by equation (3), and the subsequent tap coefficient sum of squares circuit section 23 calculates a value equivalent to the residual power in the subsequent stage using equation (4). These values are compared.

Then, the front stage power level H1 and the rear stage power level H
If Hl>a2H2 in a state where both the increase and decrease are small, it is determined that the echo path is in a fluctuating state, and ADFII is controlled to perform normal adaptive control.

In addition, when the amount of increase in the subsequent stage power level H2 is large,
If H1≦a2H2, it is determined that there is a double talk state, the adaptive control by ADFII is stopped, and the tap coefficient hi (k) immediately before this stop is held. Then, using this tap coefficient hH(k), the ADFll is made to perform the calculation of the above equation (1) to form a pseudo echo signal y (k), and this pseudo echo signal y (k) becomes the echo signal y (k).
is subtracted from.

Furthermore, when leaving the double talk state, the determination circuit 24 sets the set value RES(th) and the residual level LRES.
are compared, and when L RES (th) > L RES, adaptive control by ADFII is restarted.

As described above, by focusing on the physical properties of the tap coefficient hi (k) and comparing and determining the magnitudes of the sum of the two tap coefficients in the front stage and the sum of the two tap coefficients in the latter stage, it is possible to determine whether there is a double talk state or not due to echo path fluctuation. It becomes possible to accurately determine whether the state is the same or not.

As a result, accurate adaptive control can be performed by ADF II, and the performance and reliability of the acoustic echo canceller can be improved.

In this embodiment, as a means for determining the difference in the residual power, the front-stage tap coefficient 2 Towa circuit section 22 and the rear-stage tap coefficient square sum circuit section 23 are used, and the tap coefficients are squared and all of them are calculated. Although the value corresponding to the residual power is detected by summing the value, the present invention is not limited to this. Instead of the preceding stage tap coefficient 2 to sum circuit section 22 and the subsequent stage tap coefficient square sum circuit section 23, a value corresponding to the residual power is detected. A circuit that directly detects the size of the latter stage may be provided to determine whether or not there is double talk.

[Effects of the Invention] As described in detail above, according to the present invention, there is a front-stage power detection means for detecting the magnitude of the residual power in the front stage, a rear-stage power detection means for detecting the magnitude of the residual power in the rear stage, and each power detection unit. Comparison/judgment means for comparing output values from the means and determining a double talk state when the rate of increase in the latter stage of the residual power is large; Since it is possible to judge whether the state is a double talk state or an echo path fluctuation state, it is possible to accurately judge whether it is a double talk state or an echo path fluctuation state.As a result, it is possible to perform accurate adaptive control using an adaptive filter, and the acoustic The performance and reliability of the echo canceller can be improved.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the overall configuration of an acoustic echo canceller according to the present invention, FIG. 2 is a block diagram showing the overall configuration of a conventional acoustic echo canceller, and FIG. 3 is a block diagram showing the internal configuration of an ADF. FIG. 4 is a characteristic diagram showing a measurement example of an impulse response representing acoustic coupling between a speaker and a microphone and a corresponding residual power curve. 11...ADF, 12...Adder, 21...AD
F control circuit, 22...previous stage tap coefficient 2-sum sum circuit section,
23... Post-stage tap coefficient 2 Towa circuit section, 24... Judgment circuit. Measurement example of induction response and characteristic diagram without residual power Figure 4

Claims (1)

[Claims] In an acoustic echo canceler that combines a pseudo echo signal and an acoustic echo signal to remove an acoustic echo and stops adaptive control by an adaptive filter when in a double talk state, Of the residual power, the output value from each power detection means is compared between the front stage power detection means that detects the magnitude of the previous stage and the rear stage power detection means that detects the magnitude of the latter stage, and the increase rate of the residual power in the latter stage is determined. 1. An acoustic echo canceller comprising: comparison and determination means for determining a double talk state when .